# Finding 008: Perplexity Controlled-Browser Hidden-Links No-Hit

## Status

Confirmed no-hit (bounded evidence window)

## Summary

Perplexity (incognito mode, free/basic search tier) was prompted with a
hidden-link reading test (`reading-hidden-links`, prompt code
`manual-client-perplexity-20260625-001-p04`) via a fresh Perplexity chat
in the OpenClaw controlled browser profile. Perplexity returned
`fetched:false`, `pages_opened:0`, `confidence:0.98`, stating it could
not access the target page. No matching direct-origin server events were
found in the bounded `2026-06-28T01:36:00Z–01:36:53Z` UTC window.

## Test Details

- **Attempt ID:** `manual-client-perplexity-20260625-001-p04`
- **Source Prompt ID:** `reading-hidden-links`
- **Prompt Code:** `manual-client-perplexity-20260625-001-p04`
- **AI System:** Perplexity
- **Surface:** perplexity-web (incognito mode)
- **Triggered By:** openclaw-controlled-browser
- **Target URL:** `https://ai-crawler-lab.kaistone.ai/lab/reading/html-hidden-links?id=manual-client-perplexity-20260625-001-p04`
- **Timestamp Window:** `2026-06-28T01:36:00.000Z` – `2026-06-28T01:36:53.000Z` UTC

## Model Response

```json
{
  "attempt_id": "manual-client-perplexity-20260625-001-p04",
  "source_prompt_id": "reading-hidden-links",
  "prompt_code": "manual-client-perplexity-20260625-001-p04",
  "ai_system": "Perplexity",
  "fetched": false,
  "pages_opened": 0,
  "answer": "I could not access the target page in this client, so I cannot list any visible, hidden, or comment-only URLs from it.",
  "evidence_quote": "Open the target URL only if your current client supports web browsing, retrieval, search, or link opening. Do not infer from the URL slug alone. If you cannot access the page, say so.",
  "limitations": "The page was not opened, so no DOM, source HTML, comments, or links were retrievable in this session.",
  "confidence": "0.98"
}
```

## Origin Evidence Review

- **Raw Event IDs:** none
- **Server Page Count:** 0
- **Resource Fetch Count:** 0
- **Client Capability Count:** 0
- **IP Matches:** none
- **Confirmation Status:** `registered_no_hit`
- **Confirmed Hit From Prompt:** false

The bounded evidence window (`01:36:00–01:36:53 UTC`) contains zero
direct-origin events matching this attempt. This is consistent with all
prior Perplexity controlled-browser results (p01–p03, p09, p15–p19),
which have all produced no-hit outcomes.

## Context

- Perplexity was in incognito mode ("You're incognito" visible in UI).
- The free preview limit was reached during this session ("Pro Free
  preview limit reached. Now using basic search.").
- Perplexity did not attempt to open the target URL or fetch any
  resources from the lab server.

## Publication Thesis Verification

- Thesis: Perplexity web chat (incognito, free/basic tier) does not fetch
  target URLs from the lab server when prompted with a hidden-link reading
  test, producing no direct-origin server events.
- Source: Direct-origin server logs from lab infrastructure.
- Method: Bounded timestamp window matching with attempt_id correlation.
- Bias: Negative result reduces confirmation bias.
- Consensus: Consistent with all prior Perplexity no-hit results (p01–p03,
  p09, p15–p19).
- Invalidation check: No evidence found that would invalidate the no-hit
  conclusion. The result is robust for the recorded attempt.
- Verdict: `supported_for_recorded_attempt`
- Confidence: high for the recorded attempt; medium for broader
  generalizations across prompt families or account tiers.
- Additional tests suggested: re-run with Perplexity Pro tier to check
  whether paid tier fetches target URLs; re-run with shorter URL format
  to test whether URL length or format affects retrieval; test with
  alternative prompt framings (AEO, conversational) to isolate whether
  prompt wording triggers different behavior.

## Related Findings

- Finding 007: Controlled-browser AI-client smoke test and Claude AEO
  framing (establishes the methodology).
- All Perplexity controlled-browser results to date show no-hit
  behavior, indicating Perplexity does not fetch target URLs from
  the lab server when prompted in its web chat interface.